AI hospital coding was associated with an estimated $942 million in additional U.S. healthcare spending over 2024 and 2025, according to a new Blue Cross Blue Shield Association analysis. The insurer group says automated documentation tools are identifying more billable secondary conditions even when related treatment does not increase.

 

The September 24 findings turn a familiar AI efficiency claim into a harder accounting question: whether exhaustive coding saves administrative work while shifting higher costs to insurers, employers and patients. The study is payer-backed, however, and its claims-based method identifies an association rather than proving that software alone caused every added dollar.

 

The Blue Cross analysis highlights three changes:

  • $942 million in estimated additional spending
  • Nearly 70% tied to secondary diagnoses
  • No matching increase in related treatment

 

Further Reading

 

AI Hospital Coding Drives the $942 Million Estimate

Blue Cross Blue Shield Association based the estimate on de-identified claims data from its member companies, which collectively cover about one in three Americans. It compared the frequency of complex diagnoses and the spending attached to them as hospitals expanded use of software that reviews records for coding opportunities.

 

The association said more than 60% of hospital systems now use AI-assisted coding tools. These systems can scan clinical notes, lab results and other records, then suggest diagnoses that may support a higher reimbursement category or add charges for secondary conditions.

 

About $653 million of the estimated increase came from secondary diagnoses, conditions documented alongside the main reason for an admission. That concentration matters because secondary conditions can raise the apparent complexity of a case and change how much an insurer pays without altering the primary procedure.

 

For major bowel surgeries, Blue Cross reported a 55% rise in documented partial bowel blockages and a 33% increase in metabolic acidosis between the first quarter of 2023 and the fourth quarter of 2025. The group said it did not observe corresponding increases in care associated with those conditions.

 

More Complete Records or More Aggressive Billing

AI coding systems do not need to invent a diagnosis to increase a bill. Their commercial value often comes from finding details that busy clinicians or human coders might overlook, standardizing documentation and matching the record to the most specific available billing code.

 

That can improve accuracy when a patient genuinely has several conditions. It can also create a financial incentive to document every defensible complication, even when the condition had little effect on treatment. The disagreement is therefore not simply about whether a code is technically supported, but whether the payment reflects meaningful clinical complexity.

 

Blue Cross pointed to anemia as one example. Diagnoses increased without a comparable rise in blood transfusions, which the association used as evidence that coding intensity was changing faster than care. A lack of added treatment does not prove a diagnosis was invalid, but it gives insurers a reason to audit the pattern.

 

The analysis extends work the association published in March. That earlier research estimated that AI-linked coding patterns may have been associated with about $663 million in inpatient spending and at least $1.67 billion in outpatient spending, showing that the dispute reaches beyond a single procedure or billing category.

 

Hospitals and Insurers Enter an AI Audit Cycle

Hospitals are adopting automated coding because clinical records are long, reimbursement rules are complex and missed documentation can mean lost revenue. Insurers are responding with their own models that flag unusual diagnosis combinations, compare providers and select claims for human review.

 

This creates an AI-versus-AI audit cycle. A hospital system uses software to identify the most complete payable code, while a health plan uses separate software to decide whether that code should be challenged. Administrative work may become faster on each side even as the total volume of disputes grows.

 

Patients can feel the consequences through deductibles, coinsurance and premiums, even when they never interact with the coding software. Employers and public programs also absorb higher spending if coding intensity raises the measured cost of hospital care without a comparable improvement in outcomes.

 

The findings also complicate the market for ambient clinical scribes. Tools that transcribe visits and draft notes can reduce physicians' paperwork, but richer notes give downstream coding systems more material to classify. A product can save clinician time while increasing the amount ultimately billed.

 

Evidence Needed to Separate Accuracy From Upcoding

The analysis should not be read as a neutral clinical trial. Blue Cross member companies pay hospital claims and have a financial interest in limiting coding-driven reimbursement increases. The association released aggregate findings rather than a peer-reviewed causal study, so independent researchers will need access to methods and data to test the estimate.

 

Hospitals, for their part, need to show whether newly documented conditions changed monitoring, medication, length of stay or other care decisions that claims data may not capture cleanly. Audits should distinguish a genuinely missed diagnosis from a technically valid but clinically marginal code.

 

Regulators and payers can also require vendors to log which diagnoses were suggested by software, which were accepted by clinicians and what evidence supported each code. That audit trail would make it easier to measure false positives and compare AI-assisted records with cases coded by people alone.

 

The result is not a verdict that hospital coding AI is fraudulent. It is a quantified warning that automation can redistribute costs as well as reduce labor. The next test is whether hospitals and insurers can agree on evidence standards before automated billing and denial systems make those disputes harder to unwind.